Objective Large- diameter rotary parts play a crucial role in various applications, and non-destructive testing (NDT) is essential for preventing catastrophic failures and reducing economic losses. Conventional X-rays cannot penetrate these parts due to their substantial size and high- density materials. Therefore, an electron linear accelerator is employed as the X- ray source, typically at angles not exceeding 30 degrees . Existing digital radiography (DR) systems and motion detection enable full- section scanning of these objects. However, such methods are time-consuming and costly, particularly when precise rotation of the object is required. Given the focus on detecting defects in the shell areas, such as tangentially distributed strip defects, it is advisable to perform localized computed tomography (CT) on the shell. To address this need, we have developed a new imaging method based on the existing accelerator imaging system, aimed at enabling localized CT of large- diameter rotary part shells. This method has significant practical application value. Methods To meet the detection requirements of large- diameter rotary parts, we propose a multi- scan local source translation CT (mL-STCT) method. In mL-STCT, the detector and X-ray source are offset on opposite sides of the object. During scanning, the detector and object remain fixed, while the X-ray source translates parallel to the detector to collect projection data from various angles. The detector's center is consistently aligned with the shell area under inspection. The mL-STCT involves scanning the estimated defect region multiple times by rotating the object and adjusting the positions of the X-ray source and detector. The rotation angle and other scan parameters are determined by analyzing the projection data distribution in Radon space. Due to the limited- angle and truncation issues inherent in the mL- STCT method, the simultaneous iterative reconstruction technique (SIRT) is employed for image reconstruction. Comparative experiments are conducted using local source translation CT (L-STCT) and tangential CT (TCT) methods. Reconstructed image quality is accessed using three quantitative indicators: root mean square error (RMSE), information fidelity criteria (IFC), and structural similarity index (SSIM). The method's feasibility is validated through both numerical simulations and physical experiments. Results and Discussions We conduct simulation experiments to compare different projection data distributions and imaging methods using the phantom shown in Fig. 4. Both subjective image evaluations (Figs. 5-6) and quantitative evaluation indexes (Table 3) demonstrate that continuous projection data and increased scanning times achieve high- quality image reconstruction. Under the same magnification, detector size, and imaging parameters, mL-STCT with continuous projection outperforms L-STCT, TCT, and mL-STCT with discontinuous projection in recovering defect structures with clearer and more complete outlines (Fig. 7). We also analyze the grayscale values along the line passing through the defect center, comparing the grayscale distribution curves of true and reconstructed images of different scanning methods. The results (Fig. 8) show that the grayscale distribution for mL-STCT with continuous projection is closer to the true image. Quantitative evaluation indexes (Table 4) further confirm that mL-STCT with continuous projection produces more accurate reconstruction images. A simulated mL-STCT experimental platform (Fig. 10) is built to validate the effectiveness of this imaging method. The reconstruction results (Figs. 11-12) demonstrate that strip defects are clearly visible, and hole defects retain a relatively complete structure. Based on the results of these simulated physical experiments, an actual detection imaging experiment is conducted. The results of the accelerator CT experiments (Fig. 14) show that mL-STCT can effectively detect defects in the shells of large- diameter rotary parts. Under the same range of projection angles, the projection data for mL-STCT increases with the number of scans, while TCT only shows an increase within a certain projection angle range. Consequently, mL-STCT demonstrates superior structural recovery compared to TCT when the projection angle range exceeds the maximum coverage angle achievable by TCT. Conclusions In response to the detection needs of large- diameter rotary parts, we propose an mL-STCT imaging method based on an existing accelerator imaging system. By establishing the mL-STCT geometric model and analyzing multi- scan projection data distribution in Radon space, we determine the critical rotation angle per scan, considering the geometric relationship involved. Simulation results under various projection data distributions demonstrate that continuous projection data yields high- quality image reconstructions. Both simulation and physical experiments show that, compared with L- STCT, TCT, and mL-STCT with discontinuous projection, mL-STCT with continuous projection provides superior recovery of defect structures, producing clearer and more complete outlines. Accelerator CT experiments further confirm that mL-STCT enables effective local CT imaging of large- diameter rotary part shells. The method is easy to implement due to its simple mechanical structure, and the required rotation angle accuracy per scan is relatively low. Projection data at various angles can be obtained through localized multiple scans of large- diameter rotary parts, even when the X-ray beam angle and detector size are limited. However, the reconstruction images are prone to limited- angle artifacts due to the projection coverage angle being less than 180 degrees.Future work will focus on optimizing the optimization algorithm to achieve higher- quality imaging and better meet real- world detection requirements.
